Evaluating ScaleMP virtual shared memory architecture for use in HPC

ScaleMP has provided vSMP Foundation, a new technology for large shared memory systems which could change how some mid-tier scientific computing applications are supported. vSMP aggregates multiple x86 servers into a single high-end large memory system and operating system, representing an alternative to traditional SMP or NUMA architectures that often require expensive and specialized hardware to accomplish a similar task. 

This project looks to evaluate the vSMP system on FutureGrid's new "echo" computational resource.  We will evaluate vSMP first based on traditional HPC benchmarks usually run on distributed memory systems such as MPI-enabled clusters. Next, we will look at benchmarks such as SPEC OpenMP and PRASEC to evaluate the system's overall scalability related to multi-threaded applications. This work will also look to support novel bioinformatics applications on the new large scale shared memory system, including de-novo genome assembly and new graph algorithms.